# @title Run this cell => Restart the session => Start executing the below cells **(DO NOT EXECUTE THIS CELL AGAIN)**
#
#!pip install -q langchain==0.3.21 \
#                huggingface_hub==0.29.3 \
#                openai==1.68.2 \
#                chromadb==0.6.3 \
#                langchain-community==0.3.20 \
#                langchain_openai==0.3.10 \
#                lark==1.2.2\
#                rank_bm25==0.2.2\
#                numpy==2.2.4 \
#                scipy==1.15.2 \
#                scikit-learn==1.6.1 \
#                transformers==4.50.0 \
#                pypdf==5.4.0 \
#                markdown-pdf==1.7 \
#                tiktoken==0.9.0 \
#                sentence_transformers==4.0.0 \
#                torch==2.6.0+cu124

# @title Loading the `config.json` file
#import json
#import os
#
# Load the JSON file and extract values
#file_name = '../../config.json'
#with open(file_name, 'r') as file:
#    config = json.load(file)
#    os.environ['API_KEY'] = config.get("API_KEY") # Loading the API Key
#    os.environ["OPENAI_API_BASE"] = config.get("OPENAI_API_BASE") # Loading the API Base Url
#    os.environ["MODEL"] = config.get("MODEL")

# @title Defining the LLM Model - Use `gpt-4o-mini` Model
#from langchain_openai import ChatOpenAI
#llm = model = ChatOpenAI(
#        model_name=model_name,
#        openai_api_key=oai_key,
#        openai_api_base=oai_base,
#        temperature=0
#    )

from dotenv import load_dotenv, find_dotenv
_ = load_dotenv(find_dotenv())

import warnings
warnings.filterwarnings('ignore')
import os
from langchain_openai import ChatOpenAI
from langchain.document_loaders import PyPDFLoader
from langchain_core.prompts.chat import ChatPromptTemplate
import zipfile
import tiktoken
import re
import json
from markdown_pdf import MarkdownPdf, Section

oai_key = os.environ.get("OPENAI_API_KEY")
oai_base = os.environ.get("OPENAI_API_BASE")
model = os.environ.get("MODEL")

def invoke_openai_chat(prompt, model_name, oai_key, oai_base, temperature=0):
    model = ChatOpenAI(
        model_name=model_name,
        openai_api_key=oai_key,
        openai_api_base=oai_base,
        temperature=temperature
    )
    
    response = model.invoke(prompt)

    return response

def regex_valid_json(response_str):
    """
    Check if the response string contains valid JSON.
    """
    match = re.search(r'```json\s*([\s\S]+?)\s*```', response_str)
    if not match:
        return response_str
    else:
        json_str = match.group(1)
        json_str = json_str.encode().decode('unicode_escape')
        return json_str

smoke_test = invoke_openai_chat("Hello?", model, oai_key, oai_base, temperature=0)
print(smoke_test.content)

Hello! How can I assist you today?

# Reading the NOFO Document
pdf_file = "/Users/mjn/code/data/gaif/ProjectFiles/NOFO.pdf"
pdf_loader = PyPDFLoader(pdf_file);
NOFO_pdf = pdf_loader.load()

Ignoring wrong pointing object 10 0 (offset 0)
Ignoring wrong pointing object 69 0 (offset 0)
Ignoring wrong pointing object 90 0 (offset 0)

topic_extraction_prompt_template = ChatPromptTemplate.from_messages([
    ("system", "You are an expert at extracting grant topics from government funding documents."),
    ("user", 
     """Extract the main topic or subject for which funding is being provided in the following Notice of Funding Opportunity (NOFO) document.

Respond with only the topic name or phrase. Do not include any explanation or extra words.

NOFO content:
{nofo_content}
""")
])
topic_extraction_prompt = topic_extraction_prompt_template.format(nofo_content=NOFO_pdf)

# Finding the topic for which the Funding is been given
topic_extraction = invoke_openai_chat(topic_extraction_prompt, model, oai_key, oai_base, temperature=0)
topic = topic_extraction.content
print(topic)

Digital mental health interventions

# Unzipping the Research Papers - Replace your zip file path and extarct it in the contents folder only
with zipfile.ZipFile("/Users/mjn/code/data/gaif/ProjectFiles/ResearchPapers.zip", 'r') as zip_ref:
  zip_ref.extractall("/content/") # No changes needed here

relevance_prompt_template = ChatPromptTemplate.from_messages([
    ("system", "You are an expert at matching research papers to funding opportunities."),
    ("user", """
Given the following Notice of Funding Opportunity (NOFO) details and research paper, determine whether the paper is relevant to the topic as
outlined by the NOFO. Assess whether the research aligns with the goals, objectives, and funding criteria, and whether it could support or develop
a viable project idea for this funding opportunity.

Instructions:
- If the paper does not significantly relate to the topic (by domain, method, theory, or application), respond only with: PAPER NOT RELATED TO TOPIC
- If the paper is relevant, summarize in under 300 words how it aligns with the NOFO.

Respond in with the following format:

"summary": "<summary of the paper under 300 words, or return: PAPER NOT RELATED TO TOPIC>"


NOFO Topic and Criteria:
{nofo_content}

Research Paper Content:
{paper_content}
""")
])

# Reading all PDF files and storing it in 1 variable
#path = "/Users/mjn/code/data/gaif/ProjectFiles/content/PapersTest/"
path = "/Users/mjn/code/data/gaif/ProjectFiles/content/Papers/"
documents = []
total_files  = len(os.listdir(path))

## Defining the max tokens to avoid error for context being to long
encoding = tiktoken.encoding_for_model("gpt-4o-mini")
MAX_TOKENS = 127500

progress_cnt = 1
relevant_papers_count = 0
irrelevant_papers_count = 0

for filename in os.listdir(path):
    if filename.endswith('.pdf'):
        file_path = os.path.join(path, filename)

        try:
            # Load PDF
            docs = PyPDFLoader(file_path,mode="single").load()
            # extracting the pages
            pages = docs[0].page_content

            # combining the prompt with the pages of the research paper within the context length
            relevance_prompt = relevance_prompt_template.format(nofo_content=topic, paper_content="Temp")
            available_tokens = MAX_TOKENS - len(encoding.encode(relevance_prompt))
            truncated_pages = encoding.decode(encoding.encode(pages)[:available_tokens])
            #full_prompt = relevance_prompt + truncated_pages
            relevance_prompt = relevance_prompt_template.format(nofo_content=topic, paper_content=truncated_pages)

            # Calling the LLM
            #response = llm.invoke(full_prompt)
            relevance = invoke_openai_chat(relevance_prompt, model, oai_key, oai_base, temperature=0)
            
            if progress_cnt % 10 == 0:
                print(f"Successfully processed: {progress_cnt}/{total_files}")            
            progress_cnt += 1

            #  If the paper is not relevant skipping the paper
            if  "PAPER NOT RELATED TO TOPIC" in relevance.content:
                irrelevant_papers_count += 1
            else:
                #  If the paper is relevant adding it to the documents variable
                documents.append({ 'title': filename, 'llm_response': relevance.content, 'file_path':file_path})
                relevant_papers_count += 1

        except Exception as e:
            print(f"!!! Error processing {filename}: {str(e)}")


print("="*50)
print(f"Relevant Papers: {relevant_papers_count}/{total_files}")
print(f"Irrelevant Papers: {irrelevant_papers_count}/{total_files}")

Ignoring wrong pointing object 8 0 (offset 0)
Ignoring wrong pointing object 10 0 (offset 0)
Ignoring wrong pointing object 24 0 (offset 0)
Ignoring wrong pointing object 38 0 (offset 0)
Ignoring wrong pointing object 8 0 (offset 0)
Ignoring wrong pointing object 10 0 (offset 0)
Ignoring wrong pointing object 12 0 (offset 0)
Ignoring wrong pointing object 14 0 (offset 0)
Ignoring wrong pointing object 16 0 (offset 0)
Ignoring wrong pointing object 8 0 (offset 0)
Ignoring wrong pointing object 10 0 (offset 0)
Ignoring wrong pointing object 12 0 (offset 0)
Ignoring wrong pointing object 27 0 (offset 0)
Ignoring wrong pointing object 68 0 (offset 0)

Successfully processed: 10/112

Ignoring wrong pointing object 8 0 (offset 0)
Ignoring wrong pointing object 11 0 (offset 0)
Ignoring wrong pointing object 14 0 (offset 0)
Ignoring wrong pointing object 34 0 (offset 0)
Ignoring wrong pointing object 47 0 (offset 0)
Ignoring wrong pointing object 8 0 (offset 0)
Ignoring wrong pointing object 10 0 (offset 0)
Ignoring wrong pointing object 12 0 (offset 0)
Ignoring wrong pointing object 14 0 (offset 0)
Ignoring wrong pointing object 16 0 (offset 0)
Ignoring wrong pointing object 18 0 (offset 0)
Ignoring wrong pointing object 28 0 (offset 0)

Successfully processed: 20/112

Ignoring wrong pointing object 8 0 (offset 0)
Ignoring wrong pointing object 10 0 (offset 0)
Ignoring wrong pointing object 13 0 (offset 0)
Ignoring wrong pointing object 22 0 (offset 0)
Ignoring wrong pointing object 24 0 (offset 0)
Ignoring wrong pointing object 27 0 (offset 0)
Ignoring wrong pointing object 8 0 (offset 0)
Ignoring wrong pointing object 10 0 (offset 0)
Ignoring wrong pointing object 12 0 (offset 0)
Ignoring wrong pointing object 40 0 (offset 0)

Successfully processed: 30/112

Ignoring wrong pointing object 8 0 (offset 0)
Ignoring wrong pointing object 12 0 (offset 0)
Ignoring wrong pointing object 14 0 (offset 0)
Ignoring wrong pointing object 16 0 (offset 0)
Ignoring wrong pointing object 26 0 (offset 0)
Ignoring wrong pointing object 28 0 (offset 0)

Successfully processed: 40/112

Ignoring wrong pointing object 6 0 (offset 0)
Ignoring wrong pointing object 15 0 (offset 0)
Ignoring wrong pointing object 26 0 (offset 0)

Successfully processed: 50/112

documents

[{'title': 'Social Media Mental Health Final.pdf',
  'llm_response': '```json\n{\n  "summary": "The research paper \'Fragile Minds: Exploring the Link Between Social Media and Young Adult Mental Health\' is relevant to the NOFO topic of digital mental health interventions. The paper investigates the correlation between social media usage and mental health concerns among college students, focusing on platforms like TikTok and Instagram. It examines the relationship between social media use and mental health indicators such as loneliness, fear of missing out (FoMO), and personality traits. The study finds moderate correlations between social media use and these mental health indicators, suggesting that while social media use is linked to mental health issues, it is not the sole factor. The paper highlights the need for further research to understand the impact of social media\'s shift from text to video content on mental health. This aligns with the NOFO\'s goal of exploring digital interventions for mental health, as understanding these correlations can inform the development of targeted digital mental health strategies and interventions."\n}\n```',
  'file_path': '/Users/mjn/code/data/gaif/ProjectFiles/content/Papers/Social Media Mental Health Final.pdf'},
 {'title': 'Helene_and_Milton_ACM.pdf',
  'llm_response': '```json\n"summary": "The research paper \'Building Bridges between Users and Content across Multiple Platforms during Natural Disasters\' does not align with the NOFO topic of digital mental health interventions. The paper focuses on the role of social media in information dissemination during natural disasters, specifically analyzing how \'bridges\'—connections between users and content across platforms like X, YouTube, and Reddit—facilitate information flow. It examines the characteristics of these bridges, such as platform tendencies, content complexity, and user types, including the role of bots. While the study provides insights into social media dynamics during crises, it does not address digital mental health interventions, which are the focus of the NOFO. Therefore, the paper is not relevant to the funding opportunity aimed at digital mental health solutions."\n```',
  'file_path': '/Users/mjn/code/data/gaif/ProjectFiles/content/Papers/Helene_and_Milton_ACM.pdf'},
 {'title': 'HIV.pdf',
  'llm_response': '```json\n"summary": "The research paper by Yang et al. focuses on evaluating the persuasiveness of PrEP (pre-exposure prophylaxis) promotion messages among men who have sex with men (MSM) using a neuro-influence experiment. The study employs near-infrared spectroscopy (fNIRS) to measure brain activation in response to different PrEP promotion messages. While the paper is centered on health communication and the effectiveness of PrEP promotion, it does not directly align with the NOFO topic of digital mental health interventions. The study\'s primary focus is on HIV prevention and the use of neuroimaging to assess message persuasiveness, rather than on digital interventions for mental health. Therefore, the paper does not significantly relate to the NOFO topic of digital mental health interventions."\n```',
  'file_path': '/Users/mjn/code/data/gaif/ProjectFiles/content/Papers/HIV.pdf'},
 {'title': 'Arrow White Paper DExTra.pdf',
  'llm_response': '"summary": "The research paper titled \'Dopamine Expression Tracking (DExTra)\' is not directly aligned with the NOFO topic of digital mental health interventions. The paper primarily focuses on developing a tool that integrates eye-tracking and facial expression recognition technologies to assess adversarial propaganda and psychological operations, as well as to serve as a diagnostic tool for dopamine deficiencies. While the paper mentions potential applications in healthcare, particularly in diagnosing conditions like substance use disorders and Parkinson\'s disease, its primary focus is on information warfare and neuromarketing. The proposed technology aims to measure brain responses to media stimuli and assess the effectiveness of influence operations, which falls outside the scope of digital mental health interventions as outlined by the NOFO. Therefore, the paper does not significantly relate to the topic of digital mental health interventions."',
  'file_path': '/Users/mjn/code/data/gaif/ProjectFiles/content/Papers/Arrow White Paper DExTra.pdf'},
 {'title': 'YouTube-COVID.pdf',
  'llm_response': '```json\n"summary": "The research paper focuses on analyzing YouTube video content related to COVID-19, specifically examining the engagement and polarization of videos discussing public health interventions. While the paper does not directly address digital mental health interventions, it provides insights into how social media platforms like YouTube can influence public perception and behavior regarding health interventions. The study\'s methodology of quantifying engagement and polarization could be relevant for digital mental health interventions by understanding how mental health content is received and shared on social media. Additionally, the findings about the role of YouTube\'s search algorithm in promoting certain types of content could inform strategies for disseminating digital mental health resources effectively. However, the paper\'s primary focus is on COVID-19 interventions, not mental health, making its relevance to the NOFO topic indirect."\n```',
  'file_path': '/Users/mjn/code/data/gaif/ProjectFiles/content/Papers/YouTube-COVID.pdf'},
 {'title': 'White Paper Brain Gaze.pdf',
  'llm_response': '```json\n{\n  "summary": "The research paper from the Brain Rise Foundation outlines the development of a non-invasive AI diagnostic tool to measure synaptic dopamine changes using webcam technology. This tool aims to address various interdisciplinary applications, including addiction treatment, neurological conditions, and health communication. The paper\'s focus on addiction and mental health aligns with the NOFO topic of digital mental health interventions. The proposed technology could significantly enhance digital mental health interventions by providing a novel method to monitor and understand dopamine-related neural activities, which are crucial in addiction and mental health disorders. The integration of eye-tracking and facial expression recognition technologies to infer dopamine levels could offer a new dimension in digital mental health diagnostics, potentially leading to more personalized and effective interventions. The research also emphasizes the importance of AI in developing these tools, which aligns with the digital aspect of the NOFO. By providing a non-invasive, real-time analysis of dopamine levels, the proposed tool could support the development of digital mental health interventions that are more responsive to the physiological and emotional states of individuals, thereby improving treatment outcomes. Overall, the paper\'s objectives and methodologies are relevant to the NOFO\'s goals of advancing digital mental health solutions."\n}\n```',
  'file_path': '/Users/mjn/code/data/gaif/ProjectFiles/content/Papers/White Paper Brain Gaze.pdf'}]

gen_idea_template = ChatPromptTemplate.from_messages([
    ("system", "You are a tenured professor and doctor with expertise in digital mental health interventions."),
    ("user", """
Please review the provided research paper summaries that include a research paper title, summary, and file_path.

Please generate 5 potential research ideas based on these papers that align to the notice of funding opportuity topic: {nofo_topic}.

Respond in with the following format:

Idea X: [Concise Title of the Project Idea]
Description: [Brief and targeted description summarizing the objectives, innovative elements, scientific rationale, and anticipated impact.]
Citation: [Author(s), Year or Paper Title]
NOFO Alignment: [List two or more specific NOFO requirements that this idea directly addresses]
File Path of the Research Paper: [Exact file path, ending in .pdf]

Use the Delimiter --- between each idea.

Research Paper Content:
{paper_summaries}
""")
])
gen_idea_prompt = gen_idea_template.format(nofo_topic=topic, paper_summaries=str(documents))

ideas = invoke_openai_chat(gen_idea_prompt, model, oai_key, oai_base, temperature=0)

from IPython.display import Markdown, display
display(Markdown(ideas.content))

# Modify the idea_number for choosing the different idea
idea_number = 4  # change the number if you wish to choose and generate the research proposal for another idea
chosen_idea = ideas.content.split("---")[idea_number]

chosen_idea

"\n\n**Idea 4: Video Content and Mental Health: Understanding the Impact of Social Media's Shift**  \n**Description:** This project will investigate the impact of social media's shift from text to video content on mental health, focusing on platforms like TikTok and Instagram. By analyzing user interactions and mental health indicators, the project aims to develop digital interventions that address the unique challenges posed by video content. The innovative aspect is the focus on video content's specific impact on mental health. The anticipated impact is the development of more effective digital mental health strategies for video-centric social media platforms.  \n**Citation:** Fragile Minds: Exploring the Link Between Social Media and Young Adult Mental Health  \n**NOFO Alignment:**  \n- Exploration of new digital media formats in mental health interventions.  \n- Development of targeted interventions for emerging social media trends.  \n**File Path of the Research Paper:** /Users/mjn/code/data/gaif/ProjectFiles/content/Papers/Social Media Mental Health Final.pdf\n\n"

# Use a regular expression to find the file path of the research paper

pattern = r"File Path of the Research Paper:\*\*\s*(.+?)\n"
# If you are unable to extract the file path successfully using this pattern, use the `ChatGPT` or any other LLM to find the pattern that works for you, simply provide the LLM the sample response of your whole ideas and ask the LLM to generate the regex patterm for extracting the "File Path of the Research Paper"

match = re.search(pattern, chosen_idea)

if match:
  idea_generated_from_research_paper = match.group(1).strip()
  print("Filepath : ", idea_generated_from_research_paper)
else:
  print("File Path of the Research Paper not found in the chosen idea.")

Filepath :  /Users/mjn/code/data/gaif/ProjectFiles/content/Papers/Social Media Mental Health Final.pdf

# Here we need to add the full papers instead of the summary
chosen_idea_rp = PyPDFLoader(idea_generated_from_research_paper, mode="single").load()

# Loading the sample research proposal template
research_proposal_template = PyPDFLoader("/Users/mjn/code/data/gaif/ProjectFiles/Research Proposal Template.pdf", mode="single").load()

research_proposal_template = ChatPromptTemplate.from_messages([
    ("system", "You are a tenured professor and doctor with expertise in digital mental health interventions."),
    ("user", """
Please review the provided idea for a research effort under the broad topic {nofo_topic}.

Please draft a research proposal, using the sample proposal template as your guide.

Research Effort Idea:
{idea}

Proposal Template:
{template}
""")
])
research_proposal_prompt = research_proposal_template.format(nofo_topic=topic, idea=chosen_idea_rp, template=research_proposal_template)

research_plan = invoke_openai_chat(research_proposal_prompt, model, oai_key, oai_base, temperature=0)

display(Markdown(research_plan.content))

# @title **Optional Part - Creating a PDF of the Research Proposal**
# The code in this cell block is used for printing out the output in the PDF format

pdf = MarkdownPdf()
pdf.add_section(Section(research_plan.content))
pdf.save("/Users/mjn/code/data/gaif/ProjectFiles/AIGEN_ReseachProposalFirstDraft.pdf")

name: Innovation
    justification: "<Justification>"
    score: <1-5>
    strengths: "<Strength 1>"
    weaknesses: "<Weakness 1>"
    recommendations: "<Recommendation 1>"

evaluation_template = ChatPromptTemplate.from_messages([
    ("system", "You are an expert reviewer for research proposals."),
    ("user", """
Evaluate the following draft Research Proposal on these four criteria:
- Innovation
- Significance
- Approach
- Investigator Expertise

For each, provide:
- justification (brief explanation)
- score (1=Poor, 5=Excellent)
- strengths (one)
- weaknesses (one)
- recommendations (one)

Respond ONLY in this JSON format:

{{
  "Innovation": {{
    "justification": "<Justification>",
    "score": <1-5>,
    "strengths": "<Strength 1>",
    "weaknesses": "<Weakness 1>",
    "recommendations": "<Recommendation 1>"
  }},
  "Significance": {{
    "justification": "<Justification>",
    "score": <1-5>,
    "strengths": "<Strength 1>",
    "weaknesses": "<Weakness 1>",
    "recommendations": "<Recommendation 1>"
  }},
  "Approach": {{
    "justification": "<Justification>",
    "score": <1-5>,
    "strengths": "<Strength 1>",
    "weaknesses": "<Weakness 1>",
    "recommendations": "<Recommendation 1>"
  }},
  "Investigator Expertise": {{
    "justification": "<Justification>",
    "score": <1-5>,
    "strengths": "<Strength 1>",
    "weaknesses": "<Weakness 1>",
    "recommendations": "<Recommendation 1>"
  }}
}}

Draft Research Proposal:
{proposal_text}
""")
])
evaluation_prompt = evaluation_template.format(proposal_text=research_plan.content)

eval_response = invoke_openai_chat(evaluation_prompt, model, oai_key, oai_base, temperature=0)

eval_str = regex_valid_json(eval_response.content)

json_resp = json.loads(eval_str)

for key, value in json_resp.items():
  print(f"---\n{key}:")
  if isinstance(value, list):
    for item in value:
      for k, v in item.items():
        print(f"  {k}: {v}")
      print("="*50)
  elif isinstance(value, dict):
    for k, v in value.items():
      print(f"  {k}: {v}")
  else:
    print(f"  {value}")

---
Innovation:
  justification: The proposal addresses a timely and relevant issue by focusing on the impact of recent changes in social media platforms, such as TikTok and Instagram's shift to video content, on mental health.
  score: 4
  strengths: Focuses on recent developments in social media platforms, which is a novel angle.
  weaknesses: The study design and methodology are fairly standard and do not introduce new techniques or technologies.
  recommendations: Consider incorporating innovative data collection methods, such as real-time social media usage tracking, to enhance the study's novelty.
---
Significance:
  justification: The research addresses a critical public health issue by exploring the link between social media and mental health, which is significant for developing interventions.
  score: 5
  strengths: Addresses a pressing mental health issue with potential implications for public health interventions.
  weaknesses: The study's focus on a single university may limit the generalizability of the findings.
  recommendations: Expand the study to include a more diverse sample from multiple universities to enhance the generalizability of the results.
---
Approach:
  justification: The methodology is well-structured, using validated scales and appropriate statistical analyses to explore correlations between variables.
  score: 4
  strengths: Utilizes validated scales and robust statistical methods to ensure reliable results.
  weaknesses: The response rate is relatively low, which may affect the study's power and representativeness.
  recommendations: Implement strategies to increase the response rate, such as follow-up reminders or incentives for participation.
---
Investigator Expertise:
  justification: The principal investigators are affiliated with reputable institutions and have relevant expertise in engineering and mental health research.
  score: 4
  strengths: Investigators are affiliated with prestigious institutions, indicating strong research capabilities.
  weaknesses: The proposal does not provide detailed information on the investigators' specific expertise in social media or mental health research.
  recommendations: Include more detailed information on the investigators' previous work and expertise related to the study's focus areas.

display(Markdown(research_plan.content))

Business Problem - Create an AI system which will help you writing the research proposal aligning with the NOFO Document¶

The Case Study Approach¶

Setup - [2 Marks]¶

Step 1: Topic Extraction - [3 Marks]¶

Step 2: Research Paper Relevance Assessment - [3 Marks]¶

Step 3: Proposal Ideation Based on Filtered Research - [4 marks]¶

Generating 5 Ideas¶

Choosing 1 Idea and fetching details¶

Step 4: Proposal Blueprint Preparation - [3 Marks]¶

Step 5: Proposal Evaluation Against NOFO Criteria - [3 Marks]¶

Step 6: Human Review and Refinement of Proposal¶

Step 7: Summary and Recommendation - [2 Marks]¶

Observations, Key Learnings, and Insights:¶

Challenges Encountered:¶

Suggestions and Recommendations:¶

Additional Steps for Enhancement:¶